First of all, your data is weird. The steady state in your model is about 2.3 while the mean of your data is 12. Regarding the positive values, note the sign convention in Dynare. The unstable version is more explicit. It says:
Thus, the posterior is actually negative (although positive values are also possible, see e.g. [Computing Log Likelihood and Optimization))